Open Data Science is a publication and community site covering data science tools, machine learning tutorials, and related news.
HN commenters link it for articles on topics like Facebook's Caffe2go, image generation from noise, and slime mould decision making.
Ah, I love slime moulds (I thought "mold" was American spelling, but apparently even Scientific American spells it "mould") - thank you for the article. It's funny to think of moulds making irrational decisions.
> slime moulds make decisions through comparison to the other available options, rather than having an intrinsic value for things
That sense of subjectivity and "thought" process to make decisions, sure sounds like consciousness to me - but then again, a stream flowing down a mountain also "makes decisions", and I'm not sure if there's a quantifiable difference.
I just read the article you linked elsewhere: How 300 Matchboxes Learned to Play Tic-Tac-Toe [0]. When we call a completely mechanical process "learning", it seems to imply some kind of subjectivity and "thinking".
It's hard for me to pinpoint, what is the difference between a human and a machine learning? The difference seems merely the level of complexity.
Discussion about consciousness usually breaks down, I think, because the root of this concept is in our belief systems and philosophical view of life. At one end of the spectrum, all of existence is "conscious" in varying degrees; at the other end, nothing is really "conscious", it can all be explained as mechanical processes. The latter used to be a "heretical" worldview, to deny the "spirit" pervading all creation. These days, I'd say the former view is the odd one out, being unscientific (without experimental evidence) and based on an iffy…
The trees were acacias (there may be other species that do this, I don't know) and the pheromone was ethylene, which is a very simple gas that's used in development of other things in plants, famously like fruit ripening but likely more (can't remember, prob bud development).
<https://asknature.org/strategy/leaves-signal-presence-of-predators/>
> If a single tree under duress is using its resources and energy to signal to other trees that have no way of directly helping the tree in question, what evolutionary advantage is being provided to the tree releasing the pheromones?
Its genetic legacy is being enhanced, if the trees it's protecting are related, which likely they are (very elegant phrasing though! upvoted just for that)
> thus reproduce more successfully", and that awareness may be a form of consciousness different than our own
Hmm. Be careful here. I need to sit down and read and understand this so please recognise I'm dumping this on you, but
"Michie’s design, called MENACE, was a large pile of matchboxes that contained a number of beads and learned to play tic-tac-toe."
...
"MENACE could never win against the perfect algorithm, but ended up drawing every time after about 90 games, making it equally as perfect"
<https://opendatascience.com/menace-donald-michie-tic-tac-toe-machine-learning/>
I dunno. Where do you draw the line? HTH anyway.
This quote caught my attention: "Caffe2go is core to the machine learning products at Facebook, together with research toolchains such as Torch. Because of its size, speed, and flexibility, we're rolling Caffe2go out across Facebook's stack."
Does this mean that Torch will forever be relegated to R&D status within Facebook while production apps will always be Caffe, similar to the two-language problem[1] that Julia aims to solve? I know that Yann Lecun created Lush and Torch7 is a spiritual successor, which was used for many years in the Lecun's NYU AI research lab[2], but am curious if this means a slow decline of Torch's relevance despite the numerous Facebook torch extensions[3] released relatively recently.
[1] https://www.opendatascience.com/conferences/odsc-east-2016-stefan-karpinski-solving-the-two-language-problem/
[2] https://www.reddit.com/r/MachineLearning/comments/25lnbt/ama_yann_lecun/
[3] https://github.com/facebook/fbcunn
The trick is to start with random "gaussian noise" - something like https://opendatascience.com/wp-content/uploads/2017/03/noise3-560x300.jpg - and then iteratively modify that image until it starts to look like the concept you want it to look like.
I find the concept of a GAN - a Generative Adversarial Network - useful.
My high-level attempt at explaining how those work is that you create two machine learning models, one that tries to create fake images and one that tries to see if an image is fake or not.
The first one says "here's an image", the second one says "that's a fake", the first one learns from that and tries again, then keep going until an image scores highly on the test.
The networks are adversarial because they are trying to outwit each other.
(I'm sure a ML researcher could provide a better explanation than I can, but that's the way I think about it.)